The current most frequently used approach is the attempt to replicate traditional forms of coding using AI, as well as generating themes from those codes, typically by comparing the codes and themes created by an AI with and equivalent human-generated analysis. I have collected over 100 articles and preprints that follow this approach, I personally do not find this to be a useful application of AI, however, because it tries to follow traditional practices rather than taking advantage of the unique strengths of AIs. In particular, humans may gain insights into data by detailed coding, but an AI has no need for this mechanical process; it can rely on its strengths for summarizing by proceeding directly to the creation of themes.

The alternative is what is becoming known as a "conversational" approach to qualitative data analysis with AI. This begins by asking for themes, and then engaging in a conversation with the AI to judge and improve the quality of those initial "candidate" themes. This takes advantage of the AI program's strength for digging deeper into its initial summaries.

I have two papers on this subject that you can download here. Both were published in the International Journal of Qualitative Methods. The first, from 2023, using the older approach of comparing AI-generated themes to themes from manual analyses. The more recent one, from 2025, illustrates the conversational approach. I would also recommend a 2025 article by Hayes, also in International Journal of Qualitative Methods, that uses a similar version of AI-based qualitative analysis. In addition, you should check out the various resources at Susanne Friese's website, https://qeludra.com.

I have an article forthcoming in The Qualitative Report, out in August, that recommends AI for three purposes: pursuing information, generating information, and testing information (through conversations, as stated by David L Morgan ). I use AI frequently and also have an article forthcoming on national policies in education, but more importantly, I use AI EVERY DAY.

Here are some concerns: 1.) Developing prompts is key. The PROPER format is one way (an acronym, one of many structured ways). It is not just asking a question! 2.) Trust but verify. AI often makes mistakes and provides incorrect responses. 3.) AI hallucinations are real and common. For example, when asking for possible literature, AI often generates entirely fictitious articles with incorrectly attributed DOIs. One sees many ResearchGate responses in the questions and answers that are AI-generated cut-and-pastes and completely miss the point. Be cautious!

Jonathan Canjura I apologize for just referring to AI in a generic fashion when I specifically meant generative AI programs such as ChatGPT, Claude, etc. To my way of thinking, natural language processing (NLP) is an entirely difference approach to using AI for qualitative data analysis. So, all of the over 100 articles I looked at used generative AI as opposed to NLP.